Exploration of questionable backbone conformations in crystallographic structure models using a structural alphabet.

More than 80% of protein structure models in the Protein Data Bank have been solved using X-ray crystallography. Despite continuous improvements in this experimental technique, crystallographic structure models may still present artifacts related to the crystallization process as well as errors introduced during model building and refinement, even in high-resolution cases. Such limitations can alter atomic or residue positions, leading to local misconformations, local or domain rearrangements, and occasionally global distortions. In this study, we developed a protocol to locate residues with questionable conformations, where conformations may be uncertain, atypical or influenced by crystallographic modeling and refinement. To do so, we started from a set of 826 nonredundant X-ray protein structure models. Each X-ray model underwent an energy-minimization step that relaxes atomic geometry by reducing potential energy. Residues that exhibited different local conformations between the X-ray and minimized models were therefore considered as having questionable conformations. To identify them, we compared the X-ray and minimized models of each protein using the HMM-SA structural alphabet. Our results revealed that over 18% of the residues in the protein set have questionable conformations in their backbone. These conformations can occur either as isolated events within the protein sequence or can form patterns. Moreover, we observed that the frequency of questionable conformations per X-ray model was independent of factors such as the date of deposition, resolution or crystal system. Analysis of the properties of residues associated with questionable conformations revealed that they do not specifically occur in flexible or accessible regions. However, there is a correlation between questionable conformations and secondary structures, with a particular overrepresentation of residues with questionable conformations in α-helices. We then further investigated questionable conformations in the structure model of ligand-free HIV-2 protease (PR2). By combining our questionable conformation-detection protocol with molecular-dynamics simulation, we demonstrated that approximately half of the questionable conformations in PDB entry 1hsi correspond to local conformations that are sparsely sampled by PR2 during the molecular-dynamics simulation or are structural outliers detected by the wwPDB report. In addition, our results suggested that these questionable conformations may affect the position of the flaps, two β-sheets forming the top of the binding site. In PDB entry 1hsi, their relative arrangement appears atypical compared with MD simulations, raising questions about the biological relevance of this conformation. To conclude, we have developed a protocol to quantify and localize questionable backbone conformations in X-ray structure models, which can affect the interpretation of structural data.